Tideway: Phase-Aware Tiered Memory Management for MoE Inference on Consumer Devices Tideway introduces phase-aware tiered memory management for running large Mixture-of-Experts (MoE) models locally on consumer PCs, addressing the memory constraints that separate consumer hardware from datacenter servers. The work targets local MoE inference to balance data privacy, API costs, and inference efficiency as LLM agents become a dominant application paradigm. Large language model LLM agents are rapidly becoming a dominant paradigm for AI applications. Recently, there is a growing demand to deploy large-scale Mixture-of-Experts MoE models locally on consumer PCs to balance data privacy, API costs, and inference efficiency. However, unlike datacenter s